HN Showcase
Back to browse
Kaeso – infrastructure for connecting AI agents to real services

Kaeso – infrastructure for connecting AI agents to real services

by devinoldenburg·Mar 6, 2026·1 point·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidSolve My ProblemShip It

OAuth plumbing layer for agents, but Anthropic's Resources and LangChain already ship this.

Strengths
  • •Encrypted token vault with auto-refresh eliminates OAuth boilerplate across 20+ providers.
  • •Unified API surface means agents don't learn new auth per integration.
  • •Audit logs and granular revocation are genuinely useful for multi-agent compliance.
Weaknesses
  • •Agent OAuth infrastructure is becoming a commodity — Anthropic tools, LangChain, Continue all bundle this.
  • •Landing page promises much but repo/docs missing; 'Coming Soon' pricing suggests pre-launch.
Target Audience

AI agent developers and SaaS platforms building multi-service integrations.

Similar To

Anthropic Resources (tool use + OAuth) · LangChain's tools and auth system · Zapier's OAuth abstraction layer

Post Description

Hi HN,

I've been experimenting with systems built around AI agents recently and ran into a recurring problem: connecting agents to real-world services is still surprisingly fragmented.

Every integration tends to require its own authentication flow, token management, permissions handling, and API logic. After a few integrations the architecture becomes more about maintaining service connections than about the agents themselves.

Because of that I started building a project called Kaeso.

The idea is to explore a unified infrastructure layer where external services can be connected once and then accessed by agents through a structured interface. The goal is to simplify the integration side so developers can focus more on agent logic rather than constantly rebuilding connection layers.

While building it, the original idea actually changed quite a bit, which I wrote about here: https://kaeso.ai/blog/redefining-kaeso

The project is still early, but I'm interested in hearing how others here approach this problem.

If you're building systems around agents or automation, do you usually implement integrations separately for each project, or do you maintain some kind of shared integration layer?

Similar Projects